What is good UX in an AI world?

Software used to do nothing until we made it do something. Now, AI products built on LLMs will, from Day 1, answer anything we ask, albeit with unknown quality, accuracy, and safety.

This poses some really challenging design questions. We’ve always thought in terms of funnels: we lead users down a path via navigation, wizards, and CTAs, hopefully resulting in conversion events that generate revenue or helping the user accomplish a Job To Be Done.

What is the UX analogue in an AI world, where the entry points to a “funnel” are as infinite as human language can invent?

This is the Amazon Alexa problem: people didn’t know what they could use it for, so they memorized a handful of incantations that work (clock/timer/alarm, weather, music, re-ordering stuff) and stopped there.¹

This is also the Tesla everything-hidden-behind-a-flatscreen problem: sometimes you just want to turn the AC down and the volume up. Physical buttons and knobs are great at this, and hiding these features behind layers of navigational touchscreens might look cooler but is objectively worse UX.²

I don’t know exactly what good looks like yet, but my hunch is that AI products should present a UX that is opinionated about how we want the user to use us, without limiting the user if they have other ideas. More guideposts and guardrails than fenced in funnels.

Lots of user research and playing with prototypes awaits; excited to share it with you all as we figure this stuff out.

The Workshop

This is a newsletter-only section where I share a half-baked idea in hopes that y’all who are smarter than me can work it out with me.

Nieman Journalism Lab interviewed 7 news audience directors about Google’s introduction of AI Overviews into search results. It’s really interesting to see the range of sophistication in the answers; some really get what’s going on at a strategic level, and some do not.

The answer I agree with the most is from New York Public Radio, which talks about how their strategy continues to be to build a direct relationship with their customers. Aggregators (like Google) naturally commoditize their supply; news in particular is nearly worthless as soon as it’s no longer new.

Subscription is the right path forward, but — as we’ve all seen from the fact that so many news sites now have a paywall or are building one — only a few of the very best will win a piece of consumers’ wallet share.

The other strategy that we’ve discussed here in the past is community. Bloomberg has a new piece about how sites like Strava, Letterboxd, and AllTrails have substantially growth their audience traffic and monthly active users as a result. When I was at Etsy back in 2008, our community was a huge and underreported growth engine for us.

In a world of Google AI Overview answers in search results, becoming a destination brand - an app on people’s home screens, a place worth paying for - is even more valuable than before.

1  It’s important to admit that a big part of Alexa’s failure is that, when we did try asking it to do more things, it usually couldn’t. They tried to launch an App Store but there was too much friction. Amazon is releasing a new Alexa, built on Anthropic’s Claude AI, which will help. But it might be too late (and the strategic benefit to Amazon too tangential).

2  AI voice controls in cars is becoming more popular as a solution to this. I recently watched a car review on YouTube that admitted the UX design of the in-dash software was so bad that telling the car what to do via voice was the best way to use it. Notably, that voice control feature required an additional monthly subscription fee. Good luck with that pricing strategy.